gsd-graph

An open source toolkit for building grounded local knowledge graphs from documentation, plans, decisions, and codebase knowledge.

Use it when developers and AI agents need to ask multi-hop questions with answers grounded in source documents without a hosted graph service.

Just launchedKnowledge Graph@opengsd/gsd-graph
Run gsd-graph
gsd-graph

$ gsd-graph init docs/

extracted: 48 documents, 214 entities, 530 relations

grounding index: local sqlite-vec · 0 cloud dependencies

Q: How do milestones connect to worktrees?

› Path: Milestone → Worktree → Verification → Commit

Grounded citations: ROADMAP.md:L42, SPEC.md:L18

Extract. Normalize. Connect. Query. Ground. Local graphs that keep AI agents grounded in truth.

Roadmap

gsd-graph direction

GSD Graph is expanding support for local embeddings, AST relation extraction, and zero-config project graphs.

View product roadmap
Package
@opengsd/gsd-graph
Surface
Local Toolkit & CLI
Best for
Grounded agent memory

Local-first knowledge

Builds and queries knowledge graphs on your local machine with zero external service lock-in.

Grounded citations

Preserves exact source lineages and document spans for every extracted entity and relation.

Multi-hop reasoning

Lets AI coding agents traverse complex relationship chains across specs, ADRs, and code.

How it works

The operating loop for gsd-graph.

  1. 01Extract relationships and entities from markdown docs and plans.
  2. 02Normalize and connect cross-document references.
  3. 03Query with multi-hop grounding for agent decision making.